Source: python-msgpack
Version: 1.2.1-3
Severity: serious
X-Debbugs-Cc: [email protected]

Dear Maintainer,

The autopkgtest invocation of the upstream unittests fails for
python-msgpack. It fails both on ci.d.n with its mixed testing/unstable
setup and also locally when freshly rebuilt and then immediately run only
against packages from unstable. I've not investigated further as to why
(or how).

//---

  65s =================================== FAILURES 
===================================
 65s _______________ test_unpacker_no_refcount_underflow_after_error 
________________
 65s 
 65s     def test_unpacker_no_refcount_underflow_after_error() -> None:
 65s         """Regression test (CVE-2026-114613): after a failed skip(), the C
 65s         extension must not drop a stale reference to a previously returned
 65s         object (use-after-free)."""
 65s         up = Unpacker()
 65s         up.feed(packb({"a": [1, 2, 3]}, use_bin_type=True))
 65s         outer = up.unpack()
 65s         inner = outer["a"]
 65s         keep = inner  # safety reference keeping the probe object alive
 65s         gc.collect()
 65s         r1 = sys.getrefcount(inner)
 65s     
 65s         # Two nested fixarray headers then an invalid byte: skip() opens 
two
 65s         # containers without writing the stack slots, then fails, and the
 65s         # cleanup must not touch the stale slot left by the previous 
unpack().
 65s         up.feed(b"\x92\x92\xc1")
 65s         with raises(Exception):
 65s             up.skip()
 65s         gc.collect()
 65s     
 65s         assert sys.getrefcount(inner) == r1
 65s         # The parser context must be reusable afterwards: the same 
operation
 65s         # fails cleanly again (no corrupt-state crash), and after 
realigning
 65s         # past the rejected byte the unpacker resumes normally.
 65s         with raises(FormatError):
 65s >           up.skip()
 65s 
 65s test/test_except.py:216: 
 65s _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
_ _ _ 
 65s msgpack/fallback.py:585: in skip
 65s     self._unpack(EX_SKIP)
 65s msgpack/fallback.py:489: in _unpack
 65s     typ, n, obj = self._read_header()
 65s                   ^^^^^^^^^^^^^^^^^^^
 65s msgpack/fallback.py:396: in _read_header
 65s     self._reserve(1)
 65s _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
_ _ _ 
 65s 
 65s self = <msgpack.fallback.Unpacker object at 0x7f5b5e5b2030>, n = 1
 65s raise_outofdata = True
 65s 
 65s     def _reserve(self, n, raise_outofdata=True):
 65s         remain_bytes = len(self._buffer) - self._buff_i - n
 65s     
 65s         # Fast path: buffer has n bytes already
 65s         if remain_bytes >= 0:
 65s             return
 65s     
 65s         if self._feeding:
 65s             self._buff_i = self._buf_checkpoint
 65s >           raise OutOfData
 65s E           msgpack.exceptions.OutOfData
 65s 
 65s msgpack/fallback.py:367: OutOfData
 65s =========================== short test summary info 
============================
 65s FAILED test/test_except.py::test_unpacker_no_refcount_underflow_after_error
 65s =================== 1 failed, 130 passed, 2 skipped in 1.77s 
===================

---//

The failure above is with Python 3.14 so this is not related to the
on-going work to add Python 3.15 as a supported interpreter (but it will
get in the way of that transition at some stage).

regards
Stuart

Reply via email to